Estimation Of Physiological And Biochemical Parameters Of Winter Wheat In Guanzhong Region By Hyperspectral Remote Sensing

Wheat is an important food crop in our country.Obtaining physiological and biochemical parameters of winter wheat in real time is very important to improve grain yield and ensure the food security.In this study,winter wheat in Guan Zhong area was selected as the research object to analyze and explore the rapid estimation method of chlorophyll and phosphorus content of winter wheat canopy leaves.Canopy spectral data and contents of chlorophyll and phosphorus were collected at four key growth stages of winter wheat.SG smoothing and spectral transformation were performed on the original canopy spectral data,and correlation between chlorophyll and phosphorus of sensitive bands,spectral index of any two bands screened by CARS algorithm was analyzed.The BP neural network model,the BP-PSO optimized neural network model and the random forest model were established based on the sensitive band combination and any two band spectral indices respectively.The accuracy of the estimation models were compared to determine the best estimation models of chlorophyll and phosphorus in different growth stages of winter wheat.Hyperspectral data were extracted by UAV hyperspectral images for analysis and modeling,and chlorophyll and phosphorus contents were inverted and mapped by remote sensing,providing technical support and theoretical basis for monitoring chlorophyll and phosphorus contents of regional winter wheat.The main research conclusions are as follows:(1)The trend of original spectral reflectance of the four key growth stages of winter wheat canopy leaves was basically the same,showing a "double valley" structure in the range of 350-700 nm,and a "red edge" phenomenon in the range of 700-780 nm,forming a stable high reflectance platform near 780-1000 nm.The posterior reflectance decreases step by step,forming a number of deepening reflection valleys.In the visible band,chlorophyll content and spectral reflectance are inversely proportional,and the decrease of phosphorus content has no obvious effect on spectral reflectance.In the near infrared band,the original spectral reflectance is directly proportional to the chlorophyll content and inversely proportional to the phosphorus content.(2)In terms of chlorophyll estimation of winter wheat canopy leaves,smoothing,first derivative and continuous wavelet transform of the original spectra were carried out,and the best spectral indices of any two bands of the three were extracted.The results showed that the correlation coefficients with chlorophyll were significantly improved.The correlation coefficients of wavelet transform spectra at booting stage,heading stage,flowering stage and filling stage reached 0.65,0.57,0.75 and-0.89,respectively,which were all better than the original spectrum and the first derivative spectrum,and compared with the other three growth stages,the correlation coefficients of filling stage were better than the other three growth stages.The sensitive bands were screened by CARS algorithm,and the dimension reduction ratio was all above 0.94.The first derivative spectrum at booting stage,the original spectrum at heading stage and the first derivative spectrum at flowering stage were modeled with the stochastic forest model with good accuracy.The continuous wavelet transform spectrum at grout stage was modeled with the BP neural network model with good estimation ability,with RPD of 2.13,2.65,2.10 and 2.89,respectively.(3)The estimation method of phosphorus content in winter wheat canopy leaves was consistent with that of chlorophyll content.Three spectral transformations were carried out and the best two spectral indices were selected according to the principle of maximum correlation.The correlation coefficients of continuous wavelet transform spectra were improved most obviously,which were increased by 48%,7%,69% and 55%,respectively,compared with the original spectral correlation after smooth.In any two spectral indices,the correlation coefficient between the combination of wavelet transform spectrum and first derivative spectrum and phosphorus content increased significantly,and the correlation coefficient of DB4-RSI combination reached the maximum value of 0.79 in grout stage.The dimensionality reduction ratio of sensitive bands screened by CARS algorithm was all above0.95.The combination of wavelet transform spectrum and BP-PSO optimized neural network model had high accuracy in phosphorus content estimation,and the RPD reached 2.16,1.63,2.16 and 1.93 at booting stage,heading stage,flowering stage and filling stage,respectively,which had good sample estimation ability.(4)The hyperspectral data of UAV images were extracted at the flowering stage of winter wheat,and the spectral indices of any two bands after spectral transformation were selected according to the principle of maximum correlation to construct the content estimation model of chlorophyll and phosphorus.The results showed that: In terms of correlation coefficient with chlorophyll content,the absolute value of the spectral index of any two bands can reach above 0.74,and the distribution of the spectral index of any two bands is mainly concentrated in the visible region.The correlation between any two bands of spectral index and canopy leaf phosphorus content is good.The absolute value of correlation coefficient is-0.61 based on the ratio spectral index of the first derivative spectrum,and the band combination is 838 nm and502 nm.The spectra and model combination of winter wheat canopy leaf chlorophyll content FDS-NDSI-BP and winter wheat canopy phosphorus content FDS-RSI-RF can estimate the samples,and the RPD is 1.61 and 1.59,respectively,which can provide a theoretical basis for the monitoring of winter wheat chlorophyll and phosphorus content in small area.